# EdgeAIStack

> Vendor-neutral engineering resource for edge AI hardware. Choose and size the right platform (Jetson, Hailo, Coral, RK3588) for a camera-AI deployment, then size compute, memory, storage, network, power and total cost. All figures are dated mid-2026; manufacturer specifications are kept distinct from EdgeAIStack planning estimates. Experimental machine-readable index — no ranking claims.

## Engines
- [Camera Stream Capacity](https://edgeaistack.ai/engines/camera-stream-capacity/): How many camera streams a Jetson handles and what breaks first — decode from NVIDIA per-codec tables with datasheet sections cited, inference, memory, network, headroom ladder, cite-this permalinks.
- [Model Memory Fit](https://edgeaistack.ai/engines/model-memory-fit/): Does this LLM/VLM/ASR model fit in this Jetson's memory at this quantisation, context and concurrency — weights, KV cache, runtime overhead, OS headroom and reserve, with headroom ladder and cite-this permalinks.
- [Jetson Configuration Checker](https://edgeaistack.ai/engines/jetson-configuration-checker/): Is this Jetson module × JetPack/L4T × install method × runtime × power mode a configuration NVIDIA ships — per-check SUPPORTED / VERSION RISK / NEEDS VALIDATION / UNSUPPORTED, the on-device verify commands with expected output, remediation and known issues, every fact sourced.
- [Benchmark Explorer](https://edgeaistack.ai/engines/benchmark-explorer/): What has actually been measured for this hardware × model × precision × runtime, by whom — matching rows with evidence class and source, the expected range per precision, a cross-hardware comparison and the class-D fallback chain when nothing was measured, with a submission form to add your own measurement.
- [Benchmark Reality Check](https://edgeaistack.ai/engines/benchmark-reality-check/): I measured X; the published number is Y — why. Compares a measured fps/latency/tokens-per-s against the Benchmark Explorer expected range for the same hardware × model × precision: verdict (within/below/above range, or no reference), gap ratio, and the likely causes ranked by fit to the gap, each with the on-device command that settles it; runs the Jetson Configuration Checker when JetPack and power mode are given.
- [Hardware Selector](https://edgeaistack.ai/engines/hardware-selector/): Rank the best-fit edge AI platform from task, cameras, resolution, power and environment.
- [System Designer](https://edgeaistack.ai/engines/edge-ai-system-designer/): Size a full deployment — compute, memory, storage, network, power, cost — with a bottleneck stress test.
- [Module Power Calculator](https://edgeaistack.ai/engines/module-power-calculator/): Estimate module TDP and thermal headroom across Jetson power modes.
- [Power Budget](https://edgeaistack.ai/engines/power-budget/): Size PoE switch and total power draw for a multi-camera deployment.
- [Storage Endurance](https://edgeaistack.ai/engines/storage-endurance/): Match drive TBW to write rate and retention for 24/7 recording.
- [Memory Estimator](https://edgeaistack.ai/engines/memory-estimator/): Check model + pipeline fit in VRAM and system RAM.
- [Network Bandwidth](https://edgeaistack.ai/engines/network-bandwidth/): Estimate bandwidth for a camera set by codec and resolution.
- [Deployment Cost](https://edgeaistack.ai/engines/edge-ai-deployment-cost/): Itemized BOM with cost-per-camera breakdown.
- [All engines](https://edgeaistack.ai/engines/): Full list including Stream Calculator, GPU Sizing, Inference Estimator and Full Deployment Planner.

## Platforms & specifications
- [Specifications reference](https://edgeaistack.ai/blog/edge-ai-hardware-specifications-reference/): TOPS, power, memory and interfaces for every indexed platform.
- [Edge AI hardware guide](https://edgeaistack.ai/blog/edge-ai-hardware-guide/): Decision framework across platforms, power, storage and cost.
- [Platform catalog (hub)](https://edgeaistack.ai/platforms/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Jetson Orin Nano Super](https://edgeaistack.ai/platforms/jetson-orin-nano-super/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Jetson Orin Nano 8GB](https://edgeaistack.ai/platforms/jetson-orin-nano/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Jetson Orin NX 16GB](https://edgeaistack.ai/platforms/jetson-orin-nx-16gb/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Jetson Orin NX 8GB](https://edgeaistack.ai/platforms/jetson-orin-nx-8gb/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Jetson AGX Orin 64GB](https://edgeaistack.ai/platforms/jetson-agx-orin-64gb/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Jetson AGX Orin 32GB](https://edgeaistack.ai/platforms/jetson-agx-orin-32gb/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Jetson Thor T5000](https://edgeaistack.ai/platforms/jetson-thor-t5000/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Jetson T4000](https://edgeaistack.ai/platforms/jetson-thor-t4000/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Hailo-8](https://edgeaistack.ai/platforms/hailo-8/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Hailo-8L](https://edgeaistack.ai/platforms/hailo-8l/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Hailo-10H](https://edgeaistack.ai/platforms/hailo-10h/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Google Coral Edge TPU](https://edgeaistack.ai/platforms/google-coral-edge-tpu/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Rockchip RK3588](https://edgeaistack.ai/platforms/rockchip-rk3588/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.
- [Neousys Nuvo-9531](https://edgeaistack.ai/platforms/neousys-nuvo-9531/): Specs, hedged mid-2026 pricing, sizing fit and decision guidance.

## Reference architectures
- [Reference architectures hub](https://edgeaistack.ai/reference-architectures/)
- [SMB 4-camera budget build](https://edgeaistack.ai/reference-architectures/smb-4-camera-budget-edge-ai/)
- [Retail 8-camera edge AI](https://edgeaistack.ai/reference-architectures/retail-8-camera-edge-ai/)
- [Smart city 16-camera traffic AI](https://edgeaistack.ai/reference-architectures/smart-city-16-camera-traffic-edge-ai/)
- [Warehouse safety AI](https://edgeaistack.ai/reference-architectures/warehouse-safety-forklift-edge-ai/)

## Guides
- [DeepStream vs Frigate](https://edgeaistack.ai/blog/deepstream-vs-frigate-edge-video-analytics-2026/)
- [Jetson Orin Nano vs Orin NX](https://edgeaistack.ai/blog/jetson-orin-nano-vs-orin-nx-2026/)
- [Jetson Orin Nano vs Hailo-8](https://edgeaistack.ai/blog/jetson-orin-nano-vs-hailo-8-2026/)
- [Jetson vs Google Coral TPU](https://edgeaistack.ai/blog/jetson-vs-coral-tpu/)
- [Best edge AI kits 2026](https://edgeaistack.ai/blog/best-edge-ai-kits-2026/)
- [Jetson power consumption](https://edgeaistack.ai/blog/jetson-orin-nano-power-consumption/)
- [Jetson power modes: 5W vs 7W vs 15W](https://edgeaistack.ai/blog/jetson-orin-nano-power-modes-5w-vs-7w-vs-15w/)
- [PoE switch sizing for 8 cameras](https://edgeaistack.ai/blog/poe-switch-power-budget-8-cameras/)
- [Best SSD for 24/7 recording](https://edgeaistack.ai/blog/best-ssd-24-7-video-recording-2026/)
- [Thermal design: heatsinks & airflow](https://edgeaistack.ai/blog/edge-ai-thermal-design-heatsinks-airflow-and-deployment/)
- [RAM sizing for edge inference](https://edgeaistack.ai/blog/ram-sizing-edge-inference/)
- [All guides](https://edgeaistack.ai/blog/)

## API & agents
- [API, MCP & OpenAPI docs](https://edgeaistack.ai/api/): Call the sizing engines from code or AI agents.
- [OpenAPI 3.1 spec](https://api.edgeaistack.ai/api/v1/tools/openapi.json): Machine-readable tool definitions (recommend_platform, design_system, calculate_streams, camera_stream_capacity, estimate_bandwidth, estimate_power, estimate_storage).

## Methodology & about
- [Methodology](https://edgeaistack.ai/methodology/): How the sizing numbers are derived.
- [Jetson power modes by module and JetPack](https://edgeaistack.ai/platforms/jetson-power-modes/): Every nvpmodel preset with mode ID, power budget, clocks, Super enablement and NVIDIA source; JSON at https://edgeaistack.ai/datasets/power-modes.json.
- [Jetson decode capacity tables](https://edgeaistack.ai/decode/): NVIDIA datasheet NVDEC/NVENC stream counts per module and codec with source, revision and verification date; JSON at https://edgeaistack.ai/datasets/decode-capacity.json.
- [Camera Stream Capacity methodology](https://edgeaistack.ai/methodology/camera-stream-capacity/): Formulas, datasheet tables, evidence classes and the confidence rule for the Camera Stream Capacity engine.
- [Model Memory Fit tree](https://edgeaistack.ai/model-fit/): 22 LLM/VLM/ASR models × 7 Jetson modules, verdict matrix, per-model facts and per-model×module breakdowns with sources; JSON at https://edgeaistack.ai/datasets/model-memory.json.
- [Model Memory Fit methodology](https://edgeaistack.ai/methodology/model-memory-fit/): Weights, KV cache, vision tower, runtime overhead, OS headroom/reserve formulas, evidence classes and the confidence rule for the Model Memory Fit engine.
- [Jetson compatibility tree](https://edgeaistack.ai/compatibility/): 9 Jetson modules × 16 JetPack releases (5.1.2–7.2.1) — supported pairs with the JetPack ↔ L4T ↔ CUDA/TensorRT/cuDNN/Ubuntu stack, install methods, nvpmodel presets, known issues and the verify plan per module × JetPack; JSON at https://edgeaistack.ai/datasets/jetpack-config.json.
- [Jetson Configuration Checker methodology](https://edgeaistack.ai/methodology/jetson-configuration-checker/): The check catalogue, status vocabulary and verdict rule, the verify-command table with expected output, the JetPack release / install method / runtime pairing / known issue tables with sources, and the research gaps.
- [Benchmark Explorer tree](https://edgeaistack.ai/benchmarks/): 567 inference benchmark rows × 21 hardware platforms × 49 model families — coverage, family × precision matrix and per-hardware×model breakdowns with sources; JSON at https://edgeaistack.ai/datasets/benchmarks-v2.json.
- [Benchmark Explorer methodology](https://edgeaistack.ai/methodology/benchmark-explorer/): Row schema, the three sources merged (catalog benchmark library, research corpus, P185 tokens/s), the dedupe rule, evidence classes C/D, the expected-range and fallback rules, the comparison rule, the confidence rule and coverage gaps.
- [Benchmark Reality Check methodology](https://edgeaistack.ai/methodology/benchmark-reality-check/): The reference rule (Benchmark Explorer expected range, B → C → D fallback), measured-value normalisation, the verdict rule and constraint status, the likelihood × fit cause-ranking rule, the full 17-cause catalogue with verify commands and sources, the Jetson Configuration Checker hand-off, the submit prefill and evidence classes.
- [Measurement Protocol](https://edgeaistack.ai/methodology/measurement-protocol/): How to measure your own board (environment, model artefact, benchmark command, warm-up and sampling) and submit it to POST /api/v1/measurements/submit for review as a class-C benchmark row.
- [About](https://edgeaistack.ai/about/)